Emulsifier peptides derived from seaweed, methanotrophic bacteria, and potato proteins identified by quantitative proteomics and bioinformatics

Global focus on sustainability has accelerated research into alternative non-animal sources of food protein and functional food ingredients. Amphiphilic peptides represent a class of promising biomolecules to replace chemical emulsifiers in food emulsions. In contrast to traditional trial-anderror enzymatic hydrolysis, this study utilizes a bottom-up approach combining quantitative proteomics, bioinformatics prediction, and functional validation to identify novel emulsifier peptides from seaweed, methanotrophic bacteria, and potatoes. In vitro functional validation reveal that all protein sources contained embedded novel emulsifier peptides comparable to or better than sodium caseinate (CAS). Thus, peptides efficiently reduced oil-water interfacial tension and generated physically stable emulsions with higher net zeta potential and smaller droplet sizes than


INTRODUCTION
Due to growing consumer demand for clean label and sustainable food ingredients (Asioli et al., 2017), the search for potent, natural replacements for chemical additives (e.g. emulsifiers), is rapidly developing. In light of the tremendous carbon footprint imposed by the food sector (Poore & Nemecek, 2018), utilization of alternative sources and protein-rich industrial side-streams, as well as zero-waste ambitions, has attracted immense attention. As proteins may furthermore be processed to release functional and bioactive peptides (Hajfathalian et al., 2018), they may be regarded as a potential vast resource of natural ingredients to both replace chemical additives while also improving nutritional quality and sustainability. Consequently, peptides from various sources of relevance in foods (e.g. dairy, plant, animal, and seafood) have been reported to display diverse functional properties (Ashaolu, 2020;Hajfathalian et al., 2018;Jafarpour, Gregersen, Gomes, Marcatili, Olsen, et al., 2020). One specific class, amphiphilic peptides, has received tremendous interest, due to the diverse functional aspects related to this physicochemical property.
Amphiphilicity is a crucial factor in peptide self-assembly and therefore also of tremendous importance in the interfacial properties of peptides (Cui et al., 2010). In foods, interfacial properties are important for functionalities such as emulsification (García-Moreno et al., 2020b) and foaming (Jafarpour, Gregersen, Gomes, Marcatili, Hegelund, et al., 2020). Fish oil-in-water emulsions are attractive as delivery systems in omega-3 PUFA-enriched foods (Jacobsen, 2015). Emulsion stability is closely related to the applied emulsifier(s), as they govern oil dispersion during production as well as physical stability during storage. Emulsifiers adsorb to the oil-water interface, forming a layer to protect emulsion droplets from aggregation (e.g., flocculation, coalescence) by providing steric hindrance and/or electrostatic repulsions (McClements, 2016).
Interfacial properties also play a key role in peptide interaction with biological membranes and therefore also relate to food preservation through antimicrobial activity (Findlay et al., 2010). In fact, these properties may coincide, when assessing the functional properties of amphiphilic peptides (Dexter & Middelberg, 2008;Enser et al., 1990). Interfacial properties of amphiphilic peptides and their adsorption characteristics at the oil-water interface are highly complex but ultimately governed by the primary structure and thus influenced by various factors such as length, charge, concentration, and particularly conformation (Cheng et al., 2010(Cheng et al., , 2014Jafarpour, Gomes, Gregersen, Sloth, Jacobsen, et al., 2020;García-Moreno et al., 2020a;García-Moreno et al., 2020b;Olsen et al., 2020). Peptides, which may also exhibit antioxidant activity, have furthermore been reported to show great potential for enhancing oxidative stability in oil-in-water emulsions (Cheng et al., 2010;García-Moreno et al., 2020a;García-Moreno et al., 2020b).
Traditionally, identification of new peptide emulsifiers from relevant food protein sources has been accomplished through a top-down approach, where a protein biomass is subjected to enzymatic hydrolysis to various degrees in a trial-and-error approach (Cheng et al., 2010;Hajfathalian et al., 2018). Emulsifier peptides may subsequently be identified through different steps of bulk functional validation, hydrolysate fractionation, and finally and very rarely peptide identification by e.g. mass spectrometry (Falade et al., 2021;Hajfathalian et al., 2018;Wu et al., 2018). Peptides with α-helix or β-strand secondary structures show facial amphiphilicity and locate themselves 4 parallel to the oil-water interface by projecting lipophilic parts to the non-polar phase and hydrophilic parts to the polar phase, (Dexter & Middelberg, 2008). Some peptides may however exhibit axial amphiphilicity by orienting their hydrophilic and lipophilic parts perpendicular to the oil-water interface, independent of the secondary structure. Building on these physical prerequisites for peptide emulsification activity, we recently identified a range of emulsifier peptides embedded in potato proteins using a conceptually new bottom-up approach, which combines quantitative proteomics and bioinformatic prediction to identify potent emulsifier peptides releasable from abundant proteins in a given biomass (García-Moreno et al., 2020a;2020b). The bottom-up approach fundamentally differs from the traditional top-down approach for bioactive peptide discovery, thereby removing the need for subsequent, labor-intensive fractionation and work-up to identify potential bioactive peptides. Application of bioinformatic prediction for in silico identification of new food bioactive peptides based on either physical model computation or artificial intelligence (e.g. machine learning and neural networks) is rapidly developing (García-Moreno et al., 2020a;García-Moreno, Gregersen, et al., 2020b;Minkiewicz et al., 2019;Olsen et al., 2020;Xu et al., 2018). The studies by García-Moreno et al. (2020a;2020b) indicated that even though the computed amphiphilic scores did not fully correlate with their structure and emulsifying activity, it was overall a good prediction of peptide emulsification potential. A similar approach has been applied on e.g. hydrolysates from industrial codfish wastestreams, where bioinformatic prediction was combined with proteomics studies in order to identify specific bioactive peptides responsible for the bulk emulsifying properties (Jafarpour, Gregersen, Gomes, Marcatili, Olsen, et al., 2020).
In this work, we applied the recently proposed bottom-up approach (García-Moreno et al., 2020a) to identify new emulsifier peptides embedded in high abundance proteins of new, potential protein 5 sources for use in foods. To illustrate the broad applicability of the approach and its potential for protein valorization, we selected two highly different protein sources, both adhering to the overall goal of sustainability. On one hand, we included a biomass from the fermentation of methanotrophic bacteria (M. capsulatus and Ralstonia sp.) currently employed for animal feed, and on the other hand, a side stream from industrial seaweed (E. denticulatum) processing and carrageenan production. Additionally, a set of peptides derived from previously verified peptide emulsifiers from potato (S. tuberosum) were included. These peptides were included as they were experimentally identified by LC-MS/MS and originated from the highly abundant storage protein patatin (García-Moreno et al. 2020a). Moreover, additional data will prove useful in understanding the structure/function relationship of emulsifier peptides. Thus, this study initially (i) characterized the methanotrophic biomass using quantitative bottom-up proteomics (BUP), and subsequently (ii) predicted potential peptide emulsifiers, with different secondary structure (α-helix, β-sheet and unordered), embedded in the most abundant proteins from a seaweed extract and a bacterial biomass using bioinformatic prediction. Moreover, (iii) emulsifying activity of the predicted peptides was validated in vitro by their effect on the oil-water interfacial tension and the physical stability of 5% fish oil-in-water emulsions during six days of storage mimicking delivery emulsions for omega-3 fatty acids. Next, (iv) multivariate data analysis (principal component analysis) was used to evaluate the importance of different physicochemical properties on peptide functionality and for selection of the most promising lead peptides. Lastly, (v) templated homology modelling was used to investigate putative interfacial structure and explain emulsifying properties of the selected peptides.  (Larsen, 2002). Seaweed and potato proteins described in this study were delivered by CP Kelco (Lille Skensved, Denmark) and KMC AmbA (Brande, Denmark), respectively, as previously reported García-Moreno et al., 2020a). The rest of reagents used were of analytical grade.

Proteomics analysis of microbial proteins by 1D SDS-PAGE and LC-MS/MS
Microbial protein was analyzed as previously described (García-Moreno et al., 2020a were downloaded and both included as protein databases. LC-MS/MS data was analyzed in MaxQuant 1.6.0.16 (Cox & Mann, 2008;Tyanova et al., 2016), as previously described (García-Moreno et al., 2020a). Briefly, protein identification was performed using a 1% FDR on both protein and peptide level, reverse sequences were used as decoys for FDR control, common contaminants were included, and quantification was done using both unique and razor peptides.

8
Matching between runs and dependent peptides options were enabled. Protein quantification was performed using iBAQ (Schwanhüusser et al., 2011) and relative protein abundance was determined by relative iBAQ (riBAQ) (Shin et al., 2013). MaxQuant output files (txt folder) along with raw LC-MS/MS data are available through the linked Mendeley Data Repository (Gregersen, 2021).

Protein selection and bioinformatics prediction of embedded emulsifying peptides
Based on quantitative proteomics studies of microbial protein, highly abundant proteins (>1%  , 2020b). Emulsifier peptides were identified based on their amphiphilicity in a given conformation at the oil-water interface: i) α-helix, ii) β-strand or iii) partly hydrophobic and partly hydrophilic with no specified secondary structure (γ-peptides). Experimental sequence coverage was determined using multiple sequence alignment in CLC Sequence Viewer 8.0 (https://www.qiagenbioinformatics.com/).

Reduction of interfacial tension (IFT) by the selected peptides
The dynamic IFT of peptides at the oil-water interface was determined using an automated drop tensiometer OCA25 (DataPhysics Instruments GmbH, Filderstadt, Germany) at 25°C. For the oil-water IFT measurement, a small drop of the peptide/sodium caseinate solutions (0.2 wt% in 10 mM sodium acetate-imidazole buffer, pH=7), buffer or MQ water solution was generated using the automated syringe into a quartz glass cuvette filled with MCT oil (WITARIX® MCT 60/40, IOI Oleo GmbH, Hamburg, Germany). The image of the pendant drop was recorded every 10 s over a period of 30 min and the drop shape was analyzed using the Young-Laplace equation as described in Yesiltas et al. (2019). Changes in the IFT (mN/m) were plotted as a function of time (min). All measurements were performed in duplicate.

Emulsion production and storage experiment
Fish oil-in-water emulsions were produced using 0.2 wt% peptide or sodium caseinate as emulsifier. 0.2 wt% peptide/sodium caseinate was solubilized in 10 mM sodium acetate-imidazole buffer (pH 7) while shaken in a water bath (50 °C, 2h) and hydrated overnight (100 rpm) at room temperature. Next day, 5 wt% fish oil was added into the aqueous phase and emulsified first with a handheld ultraturrax (Polytron, PT1200E, 18000 rpm, 30s). After pre-homogenization, sonication (Microson XL2000, probe P1) was used as a secondary homogenization at 75% amplitude (max. amplitude 180 µm) for 30 s in two passes with 1 min break. Emulsions were produced in triplicate and pH was measured before and after emulsification. Emulsions were stored for six days at room temperature in darkness.
where h T is the total height of the total emulsion and h W is the height of the less opaque phase at the bottom.
CI and OI were determined by adding 2 mL emulsion to a 4 mL measuring cylinder and observations were recorded on days 0, 1 and 6 during storage.

Droplet size distribution
Droplet size distribution was measured by laser diffraction in a Mastersizer 2000 (Malvern Instruments, Ltd., Worcestershire, UK). The emulsion was diluted in recirculating water (3000 rpm), until it reached an obscuration of 10-12%. The refractive indices of sunflower oil (1.469) and water (1.330) were used as particle and dispersant, respectively. Results are given in volume mean (D[4,3]) and surface area mean (D[3,2]) diameters. Measurements were performed in triplicate on days 0, 1, 3, and 6.

Zeta potential
The zeta potential was measured in a Zetasizer Nano ZS (Malvern instruments Ltd., Worcestershire, UK) with a DTS1070 cell at 20°C to identify the surface charge of the oil droplets, which are covered with peptides. Before analysis, the emulsion was diluted in buffer (20 μL emulsion in 10 mL buffer). The zeta potential range was set to -100 to +50 mV. Measurements were performed in triplicate on day 1.

Protein modelling and peptide visualization
For the nine best performing emulsifier peptides, protein modelling was performed as previously

Proteomics analysis of biomass from methane-based fermentation
Bottom-up shotgun proteomics analysis of the homogenized biomass resulted in identification of 1839 unique peptides distributed between 597 protein groups accounting for a total of 626 potential proteins (Table S1). This corresponds well with the level of protein complexity observed from SDS-PAGE analysis (Fig. S1). Applying SDS-PAGE and subsequent in-gel digestion not only provides visualization of the biomass proteome, but also facilitate pre-fractionation according to size, thereby allowing further depth in the proteomics analysis of the biomass. As expected, the majority of identified protein groups originated from M. capsulatus (549) accounting for 93.5% of the total sample protein (by riBAQ), while Ralstonia sp. specific or indistinguishable/conserved proteins accounted for a much lower content of the protein biomass (1.4% and 5.1%, respectively). This is in good agreement with the biomass supposedly consisting of 90% M. capsulatus and 8%

Ralstonia sp. (Unibio supplied information).
On the single protein level, 16 protein groups were identified to each constitute >1% of the relative protein content. Hereof, 15 were unique proteins from M. capsulatus while one protein group 13 contained a conserved protein between the two species (Table S2). Not surprisingly, a large proportion of the abundant proteins in the methanotrophic biomass are related to the methane metabolism (mca00680) KEGG pathway (Kanehisa & Goto, 2000).

Prediction of emulsifier peptides by bioinformatics
Abundant microbial proteins (Table S2) were selected for bioinformatics prediction of embedded emulsifier peptides. Similarly, abundant proteins from a recent study on hot-water extracts from the seaweed E. denticulatum  were included in the prediction. This was done to obtain a wide variety of peptides with potential emulsification activity and sufficient production yield. Lastly, four peptides, originating from the highly abundant potato (S. tuberosum) storage protein Patatin were included in the study. These peptides were previously experimentally identified by LC-MS/MS following tryptic digestion and are derived from known emulsifier peptides, (García-Moreno et al., 2020a). In total, 28 potential emulsifier peptides (12 seaweed, 12 microbial and 4 potato peptides) with different predicted emulsification mechanism, were synthesized by a commercial peptide supplier (Table 1). The peptides varied in their physicochemical characteristics such as potential structure at the interface, length, net charge (pH 7), isoelectric point, and amphiphilicity.
Interestingly, the majority of predicted emulsifier peptides from the microbial biomass originate from either enzyme involved in methane metabolism or GroEL chaperonin 2. As these proteins are either membrane associated (Myronova et al., 2006) or form highly complex membrane-like macrostructures (Braig et al., 1994), they are thus likely to include highly amphiphilic regions, which may in turn prove to be excellent sources of emulsifier peptides. The genome and proteome of E. denticulatum is poorly described in the literature. Nevertheless, the majority of predicted seaweed emulsifier peptides originate from proteins predicted to be extracellular (Gregersen et al., 2020).

Interfacial tension
The ability of surface active compounds to decrease the oil-water IFT in emulsions is of great importance, as this ability favors the formation of emulsified oil droplets (Matsumura & Matsumiya, 2012;McClements, 2016). Therefore, the emulsifier peptides were tested to investigate their role on decreasing the IFT between MCT oil and buffer as a function of time.
Some peptide solutions were cloudy (

Seaweed-derived peptides
IFT of seaweed peptides ranged between 8.6 and 25.4 mN/m at 30 min (Fig. 1A). 89-S-G had as high IFT as MQ water, which indicates a very low surface activity. This is contradicting its high amphiphilic score as this peptide has the highest among all (Table 1) (Table 1). Interestingly, this peptide had a solubility issue resulting in a cloudy solution when dissolved in the buffer (Table 2), which could be due to its high amphiphilicity and therefore some extremely hydrophobic regions leading to peptide aggregation and/or self-assembly due to hydrophobic interactions. Formation of nanoassemblies and macromolecular self-assembly for amphiphilic peptides in aqueous solution has been widely reported (Cui et al., 2010) and have attracted attention as e.g. potential drug delivery systems (Feger et al., 2020). Nevertheless, this insolubility apparently did not fully impair reduction of oil-water IFT and may in addition be interesting to investigate further for other applications than emulsification. It is worth noting that when solubility is an issue, undissolved peptides may accumulate at the bottom of the oil droplet weighing it down and causing shape deformation, which could result in low IFT results (e.g., even though it is not related to their surface properties).

Methanotrophic bacteria-derived peptides
IFT values at 30 min for microbial peptides varied between 11.0 and 20.6 mN/m (Fig. 1B), which is a smaller range compared to seaweed peptides (Fig. 1A). It is worth noting that, 92-U-A and 100-U-G did not reach equilibrium during 30 min of the analysis, whereas the IFT values for most of the peptides levelled off within the first 5 min. Besides these two peptides, 99-U-G reached an equilibrium around 20 min starting at 22.0 and ending at 11.8 mN/m. The slow adsorption of these peptides at the oil-water interface could be due to the long amino acid chain for these peptides (28-29 AAs), which might hinder diffusion. Furthermore, these peptides were also noted to be poorly soluble (Table 2) AAs, respectively) and thus inability to adopt a stable amphiphilic α-helix at the interface.
Moreover, both variants lack the C-terminal domain found in α-10 and α-12, which corresponds to a large proportion of the helical structure in the native protein (Fig. S2). The lack of the Cterminal domain also explains their low scores (<2) which is good agreement with the low activity.
104-P-G and 105-P-G are variants of γ1 (28 AAs), which was shown to be very good emulsifier and reduced oil-water IFT considerably (García-Moreno et al., 2020b). Superior surface-activities of 104-P-G and 105-P-G (Fig. 1C) were close to the findings of c, which reported lower IFT when γ-peptides were longer than 18 AAs, although 105-P-G is only 17 AAs. These results indicate the potential as emulsifiers for the two potato peptides. It is worth mentioning that the amphiphilic scores for these peptides were modest (3.69 and 2.45), which indicate that cumulative amphiphilicity, as calculated in EmulsiPred, does not alone appear to totally describe the emulsifying potential of peptides.

Appearance of aqueous phase and pH
19 Solubility of the peptides in the aqueous phase has an influence on the emulsifying activity of the peptides (Ralet & Guéguen, 2000). Having the peptides dissolved in the buffer before the homogenization process allows the emulsifier peptides to rapidly diffuse towards the oil-water interface projecting their hydrophobic and hydrophilic sites to the oil and water phases, respectively. Table 2 shows the solubility of the peptides. Some of the peptides (13 out of 28) had a cloudy appearance, which indicates that those peptides were not totally soluble in the buffer used, and that partial aggregation or self-assembly occurred, as described above. Peptides when dissolved in buffer had pH values ranging from 4.75 to 6.15. The variation in the pH depends on the AA profile as well as acidic impurities present in the synthetic peptides used. In addition to the influence on the pH, impurities may potentially affect the observed properties to some degree. For screening and assessment of bioactive and functional properties, peptide purity varies in the literature from >70% to >98%, depending particularly on the specific functionality (Perez Espitia et al., 2012; Thery & Arendt, 2018). As no systematic studies, to the best of our knowledge, have been performed to assess the extent nor the nature of impurities, we cannot explicitly dismiss that divergence in emulsifying properties may be observed at higher purities. Here, we adhere to recent studies where >70% was indeed regarded sufficient to evaluate peptide emulsifying properties (García-Moreno et al., 2020a;2020b;. In addition, peptides of >70% purity gives significantly more reliable data compared to e.g. peptides identified in fractionated, albeit still complex, protein hydrolysates, which is commonly encountered throughout the literature as sufficient proof of peptide activity or function. Arguably, if deep and accurate characterization of e.g. peptide molecular structure is desired, a higher peptide purity would be needed.

Creaming and oil rich phase accumulation on top 20
After the production of the emulsions, physical stability was assayed during 6 days of storage (Table 2) Several emulsions displayed oil rich phase accumulation on top at a reasonable level (OI<20%), and only one peptide, 99-U-G, resulted in an OI higher than 20% (Table 2). This could be explained by the slow adsorption of the peptides at the oil-water interface as discussed in section 3.3. This presumably resulted in coalescence of the oil droplets before they were fully covered by peptide, thus leading to larger oil droplets, which moved to the top of the emulsion. Furthermore, some of the emulsions (81-S-A, 101-U-G, and 107-P-A) had an oil layer on top. This physical destabilization phenomenon is different from OI. The existence of an oil layer indicates the accumulation of oil after the breakage of the oil-water interfacial layer, also known as 'oiling-off' (McClements, 2016). This leads to upward movement of the free oil towards the top due to its lower density than the surrounding emulsion. These samples resulted in phase separation, which infers physical instability. This is in line with the high IFT values obtained for these peptides ranging between 17.5 and 24.3 mN/m (Fig. 1). Emulsions with no observable physical instability 21 were 83-S-B, 85-S-B, 97-U-B, 104-P-G, 105-P-G, and CAS (Table 2), which were considered physically stable emulsions. Some of these peptides resulted in relatively high IFT values at 30 min. Therefore, there is no apparent relationship between physical stability and low IFT values.
Indeed, low IFT only facilitates droplet breakup during emulsification, while physical stability of emulsions is mostly determined by the potential steric hindrance and/or electrostatic repulsions provided by the peptides adsorbed at the surface of the oil droplets (Berton-Carabin et al., 2018).

Zeta potential and droplet size
Zeta potential of the emulsions produced with peptides provided information about the surface charge of the peptide interfacial layer (Table 2). Peptides provide positive charge around oil droplets below their isoelectric point and negative charge above it. Isoelectric point (pI) and net charge of the peptides at pH 7 are shown in Table 1. Table 2  During homogenization, the peptide adsorption speed determines the minimum droplet size that can be obtained (i.e., the faster the adsorption rate, the droplet disruption is facilitated and smaller droplet sizes can be achieved) (McClements, 2016). Table 2 (Table 2).
Similarly, volume weighted mean dimeter, D[4,3] (Table S3), show that even more peptides performed similar to CAS at day 0 (Table S3) Furthermore, these samples were physically stable during 6 days of storage except the emulsions stabilized with 99-U-G and 107-P-A (Table 2).

Seaweed-derived peptides
Except for two peptides (87-S-G and 91-S-G), all seaweed peptides were able to stabilize fish oilin-water emulsions (Table 2) (Table S3) indicating both polydispersity and variability between replicates for measurements.
Emulsions produced with the 83-S-B and 85-S-B displayed high physical stability based on the observations of creaming and oil accumulation as well as droplet size distribution during storage, in spite of low absolute zeta potential values (<30 mV). They had 25 and 20 AAs, respectively, which were longer than the best emulsifier β-peptides length range (13-15 AAs) suggested in a previous study (García-Moreno et al., 2020a). Although these peptides were mediocre at decreasing oil-water IFT (13.7 -17.9 mN/m at 30 min, Fig. 1), they had good solubility in buffer, good physical stability and relatively small D[3,2] values compared to CAS (Table 2) during storage. Furthermore, secondary structure of these peptides were predicted to be β-strand, which is reported to provide stiffer interfaces compared to peptides adopting α-helix conformation at the interface (García-Moreno et al., 2021)

Methanotrophic bacteria-derived peptides
Nine emulsions (out of 12) produced with microbial peptides formed an emulsion at day 0, and six of them had stable D[3,2] values during 6 days of storage. 99-U-G showed a decrease in the D [3,2] and D [4,3] values, which was presumably related to the largest OI (>20%) among all the emulsions 24 even at day 0. As mentioned previously (sections 3.3.2 and 3.4.2), 99-U-G had slow adsorption at the oil-water interface (Fig. 1B), and therefore led to an emulsion with poor physical stability with high oil index. In addition, poor solubility of the peptide in the buffer might have led to an insufficient emulsifying activity. Emulsions produced with the peptides 92-U-A and 94-U-A had severe creaming (CI>40%). Indeed, 92-U-A was intermediate at decreasing oil-water IFT (16.9 mN/m at 30 min) and had one of the lowest amphiphilic score (2.45) among other peptides as well as a low zeta potential (-10 mV). In contrast, 94-U-A was able to decrease the oil-water IFT down to 12.5 mN/m and had a high amphiphilic score of 4.68, but a low zeta potential (-22 mV) and thus inefficient electrostatic repulsion, indicates the importance of oil droplet surface charge for physical stability of emulsions.
D [3,2] values range between 0.20 and 6.59 µm for the stable emulsions produced with microbial peptides. Within these emulsions, 100-U-G had a significantly larger D[3,2] value compared to CAS, which explains its severe creaming instability (CI>40%), poor ability to decrease IFT (18.2 mN/m at 30 min). The close to zero net charge at pH 7 (Table 1) indicate insufficient electrostatic repulsion between oil droplets, as confirmed by a low zeta potential of -16 mV. In contrast, 97-U-B provided one of the smallest D[3,2], which was stable during storage. This was in line with its physical stability (Table 2) and high amphiphilic score (5.07). However, it was somewhat surprising based on the mediocre reduction of the oil-water IFT (16.5 mN/m at 30 min) and the somewhat low zeta potential of 25 mV (<30 mV). Secondary structure and length (16 AAs) of 97-U-B could be the reason for its good emulsifying activity, which was suggested in a previous study for emulsifier β-peptides with a length in the range of 13-15 AAs (García-Moreno et al., 2020a).

Potato-derived peptides 25
Only three emulsions (out of 4) could be produced when using the assayed potato peptides as emulsifier. 104-P-G and 105-P-G, which are the variants of the highly emulsifying peptide γ1 106-P-A did not form an emulsion and 107-P-A had physical instabilities after production ( Table   2). This was however expected based on amphiphilic scores <2. Although 107-P-A was able to form an emulsion with low D[3,2], it had creaming and oil layer accumulation on top of the emulsion on day 1 and displayed phase separation towards the end of the storage (Table 2). This clearly shows that high zeta potential (-47 mV) is not merely enough for obtaining a stable emulsion. Although length appears to be of importance for peptides adopting α-helical conformation at the interface, the missing C-terminal domain described in section 3.3.3 is likely also a key factor in the inferior properties observed for the truncated variants.

26
PCA showed that the first four components explained 84% of the variability in the original data (Fig 2). First principle component was described by the molecular weight and length of the peptides followed by D [4,3], zeta potential, D [3,2], and pI, explaining 36% of the variability in the original data. Second principle component was largely described by IFT followed by zeta potential and accounted for 21% of the variability in the data. In addition to the above-mentioned overall analysis, some peptides providing outstanding results in certain parameters were also selected as good emulsifiers even though they were not located at the upper-left quadrant of the biplot. On the other hand, some peptides were located closer to the upper-left quadrant; however, emulsions experienced physical instability. For example, even 27 though 94-U-A and 99-U-G performed well in decreasing oil-water IFT, they had high degree of CI or OI, indicating poor emulsifying activity. Likewise, 81-S-A, 88-S-G, 90-S-G, and 94-U-A showed high creaming instability (CI>40%) and/or resulted in phase separation during the 6 days of storage and low zeta potential (<30 mV), indicating higher likelihood of physical instability.
Therefore, these peptides were not considered as good emulsifiers. On the contrary, 85-S-B did actually have good physical stability during storage; however, it had zeta potential lower than 30 mV and was not efficient in decreasing the oil-water IFT. Therefore, it was not selected. Similarly, 83-S-B had good physical stability, small droplet size and decreased IFT which made it a good candidate as an emulsifier. 97-U-B did not decrease the IFT between oil and water sufficiently; however, it provided very good physical stability in emulsion; therefore, it was selected as one of the good emulsifier peptides. Similarly, 80-S-A and 103-U-G had small and stable droplet size, low CI (≤2) and absolute zeta potential higher than 30 mV, and thereby selected as good emulsifiers.
Verified in vitro functionality of isolated peptides provide additional knowledge about structure/function relationship of emulsifier peptides as well as insight into the potential hidden gems in food proteins. Nevertheless, this may not be directly transferable to the properties observed for a large-scale hydrolysate, as these are highly complex and contain thousands of peptides (Jafarpour, Gregersen, Gomes, Marcatili, Hegelund, et al., 2020). However, preliminary investigations on potato protein hydrolysates do show that applying a designed hydrolysis for targeted release of verified peptide emulsifiers does not only release the intended peptides but also show improved bulk functionalities of the hydrolysate when compared to hydrolysates produced with other industrially relevant proteases (manuscript in preparation).

3.6.
Protein modelling and putative peptide structure 28 Templated homology modelling has proven highly beneficial to gain insight into the secondary structure of the peptides, which affects their emulsification properties (García-Moreno et al., 2020a). For instance, we previously hypothesized that a emulsifier peptide identified by bioinformatics (γ1) was likely a facial emulsifier as it originates from a partially buried, amphiphilic surface α-helix (García-Moreno et al., 2020b). The helical conformation at the interface has subsequently been confirmed by synchrotron radiation circular dichroism (García-Moreno et al., 2021). Using the approach, we visualized the nine selected peptides within a structural model of their native proteins (Fig. 3A).
In all cases, the best fitting model, which spans the target peptide, was selected to visualize the putative peptide structure. In the case of Q60B76 (103-U-G), no satisfactory model was obtained, and thus no structural information could be extracted. Quality parameters in terms of sequence identity, GMQE (Waterhouse et al., 2018), and QMEAN (Studer et al., 2020) are summarized in Table 3 and the local quality estimates are found in Fig. S3 for the selected models. Model quality assessment is further elaborated in the Supplementary Information.
Using the modelled structures, we were able to determine the putative content of secondary structural elements for the peptides (Table 3). One of the α-type peptides, 82-S-A, (as well as the two γ-type from patatin (104-P-G and 105-P-G)), are predominantly α-helical. Furthermore, two of the β-type (86-S-B and 97-U-B) are predominantly β-strand. 80-S-A and 84-S-B contains significant amounts of both structural elements. It is widely accepted that isolated peptides may partake a very different conformation than they do in their native protein (Hanazono et al., 2018).
Nevertheless, a recent study by García-Moreno et al. (2021) found that the oil/water interfacial structure of emulsifier peptides is comparable to the native conformation, if the peptide is found in a surface exposed region of the protein. In all cases, the peptides appear to be surface exposed 29 (Fig. 3A), and the local environments can hence, to some extent, be regarded similar to the oilwater interface. Consequently, the modelled structures may be regarded as probable homologues to the peptide interfacial structure.
Considering the putative peptide structures (Fig. 3A) and the distribution of hydrophobic and hydrophilic amino acids (Fig. 3B), it is evident why the peptides were both predicted and performed as good emulsifiers. In a previous study, it was shown that secondary structure, length, and pI of peptides all influence the interfacial properties (e.g., structure, viscoelasticity, and charge) and thereby physical stability of the emulsions by offering steric hindrance and electrostatic repulsions between oil droplets (García-Moreno et al., 2021). For α-helical peptides, it is a prerequisite to have both a hydrophobic and a hydrophilic face (Eisenberg et al., 1982), which is obtained by a beneficial AA distribution along the 3.4 residues involved in each turn. AAs (Enser et al., 1990).

This type of distribution is recognized in both 80-S-
For β-strand peptide emulsifiers, the prerequisite for a hydrophobic and a hydrophilic face is alternating hydrophobic and hydrophilic amino acids (Dexter & Middelberg, 2008). This distribution is very clearly seen in for instance 86-S-B throughout the peptide and also to a very large degree in 97-U-B. Both peptides are also modelled to be almost fully, undisturbed β-strands ( Fig. 3A), comprising 86% and 81% β-strand, respectively (Table 3). In addition, their length 30 correspond very well the optimal size range for β-strand peptides (García-Moreno et al., 2020a).
The distribution of β-strands is less evident in 83-S-B and 84-S-B, although they do contain regions of alternating hydrophobic and hydrophilic amino acids (Fig. 3B).
For γ-type peptide emulsifiers, one end is hydrophobic and one end is hydrophilic, which can facilitate perpendicular interaction with the interphase (Dexter & Middelberg, 2008). This distribution is highly evident for 103-U-G, 104-P-G, and, to a lesser extent, 105-P-G (Fig. 3B). As the model for 103-U-G is particularly poor and the AA distribution does not correlate with the prerequisites for α-helical or β-strand emulsifiers, the interfacial structure is unclear and very hard to predict.

Conclusions
In this study, we successfully characterized the protein content of the biomass resulting from a mixed methotrophic fermentation process by bottom-up quantitative proteomics. Using the most abundant proteins from this and other proteomics studies, we were able to predict emulsifying peptides embedded in proteins from seaweed, methanotrophic bacteria, and potato with great success. Predicted peptides were synthesized and subjected to in vitro assays, where a high number of the predicted peptides showed better or comparable emulsifying activity to sodium caseinate in fish oil-in-water emulsions. The emulsifying activity of peptides varied based on their solubility, the ability of decreasing IFT, and structural characteristics such as length, isoelectric point, hydrophilic/lipophilic balance and putative conformation. Peptides predicted to partake α-helical structure were found to be more efficient when the length is sufficient to form helical structure, whereas peptides adopting β-strand were better emulsifiers at shorter lengths (14-16 AAs) where faster diffusion and less likelihood of aggregate formation played an important role. Out of 28 assayed peptides, nine were selected as good emulsifiers based on their in vitro emulsifying potential. All three analyzed protein sources were represented among the nine peptides, which are found in proteins constituting from 4-33% of the total protein content in industrially relevant raw materials and side-streams, thereby making them highly promising leads as large-scale obtainable food bioactives. This demonstrates the high potential and the strengths of applying a bottom-up strategy compared to traditional top-down approaches in the search for new bioactive peptides.
Furthermore, based on the predicted emulsification mechanism and putative structure from template-based modelling, the nine peptides also represent different types of peptide emulsifiers, thereby making this study a valuable contribution for a deeper understanding of what constitutes a good peptide emulsifier.  creaming.

28
**Statistical differences between storage days are shown with the letters 'a' and 'b' and the differences between samples at day 0 are 29 shown with the letters 'w', 'x', 'y', and 'z' (p<0.05).